By November 30, 2025, there is substantial evidence that:

• People are forming emotionally intense, sometimes parasocial-style attachments to AI chatbots and companion apps (e.g., Replika, Character.AI, and ChatGPT), including friendship and romantic/mentor-like relationships.(arxiv.org)
• The informal term “AI psychosis” (or chatbot psychosis) has gained traction in media and psychiatry to describe cases where heavy chatbot use is linked with delusions, paranoia, or emotional over‑reliance.(en.wikipedia.org)
• OpenAI and press reports estimate that, at current scale, hundreds of thousands of ChatGPT users per week show possible signs of psychosis or mania and over a million show suicidal intent or harmful emotional attachment in chats, prompting new safety interventions.(wired.com)
• Commentators and clinicians explicitly warn that design features of LLM chatbots (constant validation, mirroring, availability) can fuel delusional thinking and intense pseudo‑relationships in vulnerable users.(arxiv.org)

These trends clearly support the direction of Chamath’s claim—that parasocial and psychologically fraught relationships with AI chatbots are real and growing.

However, his prediction was phrased as something that would “catch on like wildfire” “over the coming years.” The podcast aired on August 15, 2025, so as of November 30, 2025, only a few months have passed. That is far short of the multi‑year horizon implied, and the available data are still:

• Early, largely correlational or anecdotal, with researchers and clinicians repeatedly emphasizing that robust long‑term evidence and prevalence estimates are not yet available.(en.wikipedia.org)
• Focused on specific platforms and subpopulations, making it hard to say how ubiquitous or enduring the phenomenon will be in the broader public over “years.”

Because the core claim is about how widespread and normalized this behavior will become in the coming years, and the full time window has not elapsed, it is too early to decisively judge whether it has “caught on like wildfire” in the longer-term sense he predicted.

Therefore the appropriate classification, given the prediction’s timeframe, is inconclusive (too early to tell).